[TargetLowering] use stable_sort to avoid nondeterminism

After 330fa7d we were seeing nondeterministic failures of
llvm/test/CodeGen/ARM/thumb-big-stack.ll, with different code being
generated in different runs.

Switching sort -> stable_sort fixes this.
It looks like the old algorithm picked the first best option, and using
stable_sort restores that behavior.

Author: sam-mccall

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

@@ -5843,9 +5843,10 @@ TargetLowering::ConstraintGroup TargetLowering::getConstraintPreferences(
     Ret.emplace_back(Code, CType);
   }
 
-  std::sort(Ret.begin(), Ret.end(), [](ConstraintPair a, ConstraintPair b) {
-    return getConstraintPiority(a.second) > getConstraintPiority(b.second);
-  });
+  std::stable_sort(
+      Ret.begin(), Ret.end(), [](ConstraintPair a, ConstraintPair b) {
+        return getConstraintPiority(a.second) > getConstraintPiority(b.second);
+      });
 
   return Ret;
 }